We investigated whether responses from different cone classes are combined before or after the nonlinearity that is responsible for generating nonlinear response components of the flicker electroretinogram (ERG). We measured the nonlinear response of the retina while systematically varying the modulation in the long-wavelength-sensitive and middle-wavelength-sensitive cones by changing the proportions of flickering 633- and 543-nm lights that compose a sum-of-sinusoids temporal waveform. We found that at high temporal frequencies the ERG responses are best accounted for by a model in which the principal retinal nonlinearity is located before the convergence of signals from the two cone classes. At low temporal frequencies the ERG signal is dominated by cone-antagonistic responses. At frequencies of 30 Hz and higher, the flicker ERG and psychophysical flicker photometry give similar estimates of the relative proportions of long- and middle-wavelength-sensitive cones. The ERG photometric null is frequency dependent, whereas the psychophysically determined ratio is much less sensitive to changes in frequency.